2. Design¶
BOM¶
This device has an extremely minimal BOM.
- 330k $\Omega$ resistor
- 220 $\Omega$ resistor
- Arduino Uno
- ADS 1015
- Nellcor™ OxiMax-compatible sensor probe
Wiring Diagram¶
Below, the wiring is presented. The green wire represents the data line, red wire represents $V_{DD} = 5V$, black wire represents ground, purple and white wires are for the LEDs.
Circuit Schematic¶
Below, the circuit schematic is presented. Similar coloring conventions have been used as the wiring diagram for readability. The resistor values were chosen according to the project specifications.
Since the all of signal processing happened digitally, let us go over each step of the code, using the below pseudocode. Detailed implementation can be found in pulseox_app.py and in commented_pulse_ox.ipynb.
Arduino Code¶
The raw code is attached to the submission, and below is the pseudocode.
- Every 5 milleseconds
- 5V to Red LED
- 0V to IR LED
- delay 5ms
- read from ADC
- 5V to IR LED
- 0V to Red LED
- delay 5ms
- read from ADC
Explanation of pseudocode design choices¶
The 5ms delay before reading from the ADC is there to ensure that there is time for the LED current to reach its intended value and thus fully emit light.
This results in a total data acquisition time of 15 milliseconds. Considering a baud rate of 115200, We can running the .sizeof() on the output string on the arduino, which returned 6 bytes. Let us say each byte is 10 bits for flexibility. This gives us 1920 outputs per second. However, we do discard several reads to account for data corruption, so let us say we get 1500 outputs per second. This gives us 0.6ms interval between each output read. Therefore, we estimate that the arduino itself is able to handle 16ms intervals between each output.
Digital Signal Processing¶
Pseudocode¶
Explanation of pseudocode design choices.¶
The data acquisition delay in the arduino was 16ms. Giving this some headroom to be 30ms, we would obtain 100 samples in about 3 seconds. The take_snapshot() function is then called approximately every 3 seconds and extracts heartrate/oxygen for the past n=200 samples (6 seconds) of pulse data. The take_snapshot() function is utilized so that we would not be printing the information too fast in a way that would be hard to comprehend by the user, and to minimize error through averaging. Considering the resting human heartrate ranges from 60bpm to 100bpm, a window of 6 seconds would get at least 6 beats and at most 10 beats. A 20 year old's running heart rate goes up to 200bpm, in which the window would contain almost 20 heartbeats. By obtaining multiple heartrate/oxygen for a given window, we can extract outliers and take the average to prevent DSP errors from confusing the user. This design is forgiving on sacrificing some data for the ease of use and readability for the user.
Within the take_snapshot() function, the DC value is extracted using a moving average filter. The parameter was chosen such that the averaging window would be approximately the length of a pulse. This allows us to extract the DC value most accurately. Since we are getting 3 ~ 5 beats in 100 samples, the averaging window was chosen to be n=20. We want our window to cover a section of a pulse. A small caveat to this method is that the output of the rolling average is smaller in length than the input array. So we simply repeat the last DC value to extend the length of the DC array. Next, the peak indices are found using scipy.signal.find_peaks. The prominence parameter was found by trial and error of visual inspection. Then oxygen concentration was calculated using the calibration curve given by Tremper and Barker. One important line of code here is checking that the number of peaks in the IR signal is equal to the number of peaks in the Red signal. Otherwise, it wouldn't make sense to calculate the ratio between the heights of the two curves. The heartrate was calculated by using the inverse of the inter-peak times of V2, which is the IR voltage. This design choice was made because the IR signal was stronger in magnitude compared to the Red signal. By using only the IR signal, we reduce logic complexity and increase reliability. The output of the take_snapshot() function is an array of oxygen levels and heartrates, with n=number of peaks elements in the arrays.
Four ColumnDataSources are maintained and updated. data, which contains the raw voltage values and time in seconds, peak_time_source, which contains the time (in milliseconds) at which the peaks occur, heartrate_source, and oxygen_source for heartrate and oxygen data. Each of these information to be maintained in a separate dictionary since they may update at different rates. We store the peak_time_source to know when to take the next snapshot (i.e. when the current time is more than 3 seconds than the last peak that was extracted)
When displaying the oxygen/heartrates, we display by averaging the array of oxygen/heartrates outputted by take_snapshot(). We exclude outliers by using the scipy.stats.trim_mean(5%) function, which slices off the left most and rightmost 5% of the array.
User Interface¶
The user interface looks as below
The heartrate and oxygen concentration level is printed in large bold letters for readability. The raw pulse traces are shown, and the plot has an automatically adjusting y range to accomodate for drift and differences in voltage values for different users. The acquire button is used to start and pause the instrument, the save button is used to save the raw traces, and the shut down button is used to shut down the app.
